Re: [PATCH 4/4] f2fs/009: detect and repair nlink corruption

[David Disseldorp <ddiss@suse.de>]

Thanks for submitting the new test. It looks good aside from a few minor
things below...

On Thu,  6 Mar 2025 16:18:09 +0800, Chao Yu wrote:

> This is a regression test to check whether fsck can handle corrupted
> nlinks correctly, it uses inject.f2fs to inject nlinks w/ wrong value,
> and expects fsck.f2fs can detect such corruption and do the repair.
> 
> Cc: Jaegeuk Kim <jaegeuk@kernel.org>
> Signed-off-by: Chao Yu <chao@kernel.org>
> ---
>  tests/f2fs/009     | 142 +++++++++++++++++++++++++++++++++++++++++++++
>  tests/f2fs/009.out |   2 +
>  2 files changed, 144 insertions(+)
>  create mode 100755 tests/f2fs/009
>  create mode 100644 tests/f2fs/009.out
> 
> diff --git a/tests/f2fs/009 b/tests/f2fs/009
> new file mode 100755
> index 00000000..8f6a3e11
> --- /dev/null
> +++ b/tests/f2fs/009
> @@ -0,0 +1,142 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2025 Chao Yu.  All Rights Reserved.
> +#
> +# FS QA Test No. f2fs/009
> +#
> +# This is a regression test to check whether fsck can handle corrupted
> +# nlinks correctly, it uses inject.f2fs to inject nlinks w/ wrong value,
> +# and expects fsck.f2fs can detect such corruption and do the repair.
> +#
> +. ./common/preamble
> +_begin_fstest auto quick
> +
> +_require_scratch

You should probably check for F2FS_INJECT_PROG and skip if not present.
Is it dependent on CONFIG_F2FS_FAULT_INJECTION? If so it'd be nice if
you could check for that too.

Flag the fix for the regression via e.g.:
_fixed_by_git_commit f2fs-tools 958cd6e

> +
> +dir=$SCRATCH_MNT

I think it'd be easier to follow if you dropped this alias and just used
$SCRATCH_MNT directly.

> +filename=$dir/foo
> +hardlink=$dir/bar
> +
> +for ((i=0;i<14;i++)) do
> +	echo "round: " $i >> $seqres.full
> +
> +	_scratch_mkfs "-f" >> $seqres.full
> +	_scratch_mount >> $seqres.full
> +
> +	if [ $i == 0 ]; then
> +		touch $filename
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`

Use "stat -c '%i'" instead of piping into awk. Also, it looks like this
is called in every round, so just move it after the elifs.

> +		nlink=0
> +	elif [ $i == 1 ]; then
> +		mkdir $filename
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		nlink=1
> +	elif [ $i == 2 ]; then
> +		mknod $filename c 9 0
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		nlink=0
> +	elif [ $i == 3 ]; then
> +		mknod $filename b 8 0
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		nlink=0
> +	elif [ $i == 4 ]; then
> +		mkfifo $filename
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		nlink=0
> +	elif [ $i == 5 ]; then
> +		socket -s $filename >> $seqres.full 2>&1 &
> +		pid=$!
> +		sleep 2
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		kill $pid >> $seqres.full 2>&1
> +		nlink=0
> +	elif [ $i == 6 ]; then
> +		ln -s $dir/empty $filename
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		nlink=0
> +	elif [ $i == 7 ]; then
> +		# orphan inode
> +		touch $filename
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		$F2FS_IO_PROG write 1 0 1 zero atomic_commit $filename 5000 >> $seqres.full 2>&1 &
> +		stat $filename >> $seqres.full
> +		rm $filename
> +		$F2FS_IO_PROG shutdown 1 $dir/ >> $seqres.full
> +		sleep 6
> +		nlink=1
> +	elif [ $i == 8 ]; then
> +		# hardlink on file
> +		touch $filename
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		ln $filename $hardlink
> +		nlink=0
> +	elif [ $i == 9 ]; then
> +		# hardlink on charactor
> +		mknod $filename c 9 0
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		ln $filename $hardlink
> +		nlink=0
> +	elif [ $i == 10 ]; then
> +		# hardlink on blockdev
> +		mknod $filename b 8 0
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		ln $filename $hardlink
> +		nlink=0
> +	elif [ $i == 11 ]; then
> +		# hardlink on pipe
> +		mkfifo $filename
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		ln $filename $hardlink
> +		nlink=0
> +	elif [ $i == 12 ]; then
> +		# hardlink on socket
> +		socket -s $filename >> $seqres.full 2>&1 &
> +		pid=$!
> +		sleep 2
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		kill $pid >> $seqres.full 2>&1
> +		ln $filename $hardlink
> +		nlink=0
> +	elif [ $i == 13 ]; then
> +		# hardlink on symlink
> +		ln -s $dir/empty $filename
> +		ino=`stat $filename | awk '/Inode:/ {print $4}'`
> +		ln $filename $hardlink
> +		nlink=0
> +	fi
> +
> +	if [ $i != 7 ]; then
> +		stat $dir/* >> $seqres.full
> +	fi
> +	echo "ino:"$ino >> $seqres.full
> +	echo "nlink:"$nlink >> $seqres.full
> +
> +	_scratch_unmount
> +
> +	$F2FS_INJECT_PROG --node --mb i_links --nid $ino --val $nlink $SCRATCH_DEV >> $seqres.full
> +	if [ $? != 0 ]; then
> +		exit
> +	fi

xfstests failures are normally triggered via golden output mismatch
instead of explicit status checks or _fail calls... Can you filter the
inject / repair output so that it ends up in the golden output for
comparison?

> +	export FSCK_OPTIONS="-f"
> +	_repair_scratch_fs >> $seqres.full
> +	if [ $? != 1 ]; then
> +		_fail "fsck can not detect and repair zero nlink corruption "$i
> +		exit
> +	fi
> +
> +	export FSCK_OPTIONS=""
> +	_check_scratch_fs >> $seqres.full
> +	if [ $? != 0 ]; then
> +		_fail "fsck hasn't fixed nlink corruption "$i
> +		exit
> +	fi
> +
> +	_scratch_mount >> $seqres.full
> +	_scratch_unmount
> +done
> +
> +echo "Silence is golden"
> +
> +status=0
> +exit
> diff --git a/tests/f2fs/009.out b/tests/f2fs/009.out
> new file mode 100644
> index 00000000..7e977155
> --- /dev/null
> +++ b/tests/f2fs/009.out
> @@ -0,0 +1,2 @@
> +QA output created by 009
> +Silence is golden